A spline approach to nonparametric test of hypothesis

We propose a test of hypothesis for the closeness of two distributions whose test statistic is asymptotically normal. The divergent is based on the estimation procedure developed in Dias (2000) using a proxy of symmetrized Kullback-Leibler distance. Simulation results show that for mixture of normal distributions this test is more powerful than Kolmogorov-Smirnov test. As an application we compare acoustic data from several languages in order to identify rhythmic classes.